Trustworthy AI LLM Scalability Risk Index (LSRI): A Cybersecurity Framework Assessing Agentic-AI Security & Software Model Supply Chain Safety Boosting AI-Generated Malware Defense & Explainability Mitigating Emerging Risks of Generative AI

Autonomous AI systems need new security approaches: Traditional cybersecurity testing methods (SAST, DAST) are insufficient for modern AI agents that can execute code and use tools in real-time. The paper argues that as AI moves beyond human oversight to autonomous multi-agent systems, entirely new security frameworks are required.

AI systems fail under stress in predictable ways: The LSRI framework identifies specific operational thresholds where autonomous AI systems break down due to heavy workloads, including increased hallucinations, data poisoning attacks, and prompt injection vulnerabilities—suggesting that security risks scale with system complexity.

Supply chain attacks on AI models are a critical vulnerability: The framework addresses threats like malicious model adapters, weight tampering, and unsafe artifacts from public repositories like Hugging Face, highlighting that AI security must extend beyond the model itself to protect the entire software supply chain.

Cryptographic verification can replace post-hoc safety measures: Rather than relying solely on training methods like RLHF (Reinforcement Learning from Human Feedback), the paper proposes using cryptographic attestation and continuous runtime verification to ensure AI systems remain trustworthy during operation.

Compliance frameworks are emerging for responsible AI deployment: LSRI aligns with established standards (NIST AI RMF, OWASP Top 10 for LLMs, ISO 42001), suggesting that industry-wide security baselines for generative AI are becoming standardized and deployable across organizations.